DEGREES OF FREEDOM REGION IN MULTI-CELL RANDOM

BEAMFORMING
Hieu Duy Nguyen, Rui Zhang, and Hon Tat Hui
ECE Department, National University of Singapore.
Email: {hieudn, elezhang, elehht}@nus.edu.sg
ABSTRACT
Random beamforming (RBF) is a practically favorable
transmission scheme for multiuser multi-antenna down-
link systems. This paper studies the asymptotic rates
achievable with RBF in a multi-cell system subject to
the inter-cell interference, by assuming that the num-
ber of users in each cell scales in a given order with the
signal-to-noise ratio (SNR). In particular, we investigate
the achievable degrees of freedom (DoF) for the sum-rate
in each cell when the SNR goes to innity, and charac-
terize the achievable DoF region for all the cells. Our
results show that to achieve the DoF-optimal transmis-
sion in a multi-cell system with RBF, the numbers of
transmit beams in all the cells need to be assigned in a
collaborative manner based on the user densities.
Index Terms Random beamforming (RBF), de-
grees of freedom (DoF), multi-cell system.
1. INTRODUCTION
In a landmark paper [1], Viswanath et al. introduced the
single-beam Opportunistic Beamforming (OBF) scheme
for a multiuser downlink system, which exploits the mul-
tiuser diversity gain and requires only partial channel
knowledge at the base station (BS). One major improve-
ment over OBF was later proposed in [2] to capture
the additional spatial multiplexing gain by transmitting
more than one random beams, thus referred to as Ran-
dom Beamforming (RBF). The achievable sum-rate with
RBF in a single cell has been shown in [2] to scale identi-
cally to that with the Dirty-Paper-Coding (DPC) scheme
assuming perfect channel knowledge at the BS as the
number of users goes to innity, for any given signal-to-
noise ratio (SNR).
Although substantial subsequent investigations and
extensions for single-cell RBF have been pursuit, there
is very few work on the performance of RBF in a more re-
alistic multi-cell setup, where the inter-cell interference
(ICI) becomes a dominant factor. In [3], the authors
proposed a collaborative transmission strategy whereby
the BS utilizes perfect channel knowledge to serve one
in-cell user with maximum ratio transmission, and em-
ploys RBF to support neighboring cells users oppor-
tunistically. In addition, the sum-rate scaling law for the
multi-cell system with RBF has been recently shown in
[4] to be similar to the single-cell result given in [2], based
on an approximation of the distribution of the signal-to-
interference-plus-noise ratio (SINR).
In this paper, we study the achievable rates of a
multi-cell RBF scheme in the regime with the number of
users per cell and the operating SNR being both large.
Specically, we assume that the number of users in each
cell scales in a predened order with the SNR, as the
SNR goes to innity. Thereby, we are able to investigate
the achievable degrees of freedom (DoF) for the sum-rate
in each cell with RBF, and furthermore characterize the
achievable DoF region for all the cells. It is worth noting
that the multi-cell downlink channel can be included in
a general interference channel setup due to the ICI, and
there have been extensive studies on the achievable DoF
of interference channels recently (see, e.g., [5],[6]). How-
ever, the achievability of the DoF in the above work is
known to rely on a transmission scheme so-called Inter-
ference Alignment (IA), which requires the perfect chan-
nel knowledge for all intra-cell and inter-cell links. In
contrast, our study focuses on the achievable DoF of a
multi-cell system with only partial channel knowledge
at the BS due to the use of RBF. In addition, the IA
scheme is in general applicable with a small number of
users in each cell [5], while our work considers the case
with a much large number of users per cell that even
scales with the SNR.
The main results of this paper are summarized as fol-
lows: We rst derive a closed-form expression of the opti-
mal achievable DoF for RBF in a single cell setup, when
the number of users scales in a given order of the SNR. It
is revealed that the number of transmit beams with RBF
needs to be properly assigned based on the user density
in order to achieve the optimal DoF. We then seek to
obtain a general characterization of the DoF region for
the multi-cell downlink system with RBF. Our results
suggest that a collaboration between dierent BSs in as-
signing their respective number of beams could achieve
better performance in terms of sum-rate and fairness.
2. SYSTEM MODEL
We consider a multi-cell system consisting of C cells,
each of which has a BS with N
T
antennas to coordinate
the transmission with K
c
single-antenna mobile stations
(MSs), K
c
1 and c = 1, , C. In the c-th cell, the
c-th BS transmits M
c
N
T
orthonormal beams and
selects M
c
from K
c
users for transmission at each time.
The received signal of user k in the c-th cell is given by
y
(c)
k
=h
(c,c)
k
Mc

m=1

(c)
m
s
(c)
m
+
C

l=1, l=c

l,c
h
(l,c)
k
M
l

m=1

(l)
m
s
(l )
m
+ n
(c)
k
, (1)
where h
(l,c)
k
C
1M
l
is the channel vector between the l-
th BS and the k-th user of the c-th cell, and it is assumed
that all elements of h
(l,c)
k
are independent and identically
distributed (i.i.d.) circularly symmetric complex Gaus-
sian random variables with zero mean and unit variance,
denoted by CN(0, 1);
l,c
stands for the signal attenu-
ation from the l-th BS to the all users of the c-th cell,
l = c;
(c)
m
C
Mc1
and s
(c)
m
are the m-th randomly gen-
erated beamforming vector and transmitted data symbol
from the c-th BS, respectively. It is assumed that each
BS has the total sum power, P
T
, i.e., Tr
_
E[s
c
s
H
c
]
_
P
T
,
where s
c
= [s
(c)
1
, , s
(c)
Mc
]
T
. It is also assumed that the
background noise n
(c)
k
CN(0,
2
), k, c.
In this study, we apply the conventional RBF [2] in
each cell. In the training phase, the c-th BS generates
M
c
orthonormal beams,
(c)
1
, ,
(c)
Mc
and uses them
to broadcast the training signals to all users in the c-
th cell. The total power of each BS is assumed to be
distributed equally over M
c
beams. Each user in the c-
th cell measures the SINR values for each of M
c
beams,
which are shown in (2) below, and feeds them back to
the corresponding BS.
SINR
(c)
k,m
=
P
T
M
c

_
h
(c,c)
k
_

(c)
m

2
__

2
+
P
T
M
c
Mc

i=1,i=m

_
h
(c,c)
k
_

(c)
i

2
+
M
l

l=1,l=c

l,c
P
T
M
l
M
l

i=1

_
h
(l,c)
k
_

(l)
i

2
_
,
(2)
where m = 1, , M
c
. The c-th BS schedules transmis-
sion to a set of M
c
users for each time by assignning its
m-th beam to the user with the highest SINR, i.e.,
k
(c)
m
= arg max
k=1, ,Kc
SINR
(c)
k,m
.
Then, the achievable sum rate in bits per complex di-
mension of the c-th cell is given by
R
(c)
sum
= E
_
log
2
_
1 + SINR
(c)
k
(c)
m
,m
__
(a)
M
c
E
_
log
2
_
1 + SINR
(c)
k
(c)
1
,1
__
. (3)
Note that (a) holds due to the homogeneous channel dis-
tribution assumed for each cell, and the negligible prob-
ability of assigning multiple beams to one user. Similar
to [5], we dene the DoF region for the multi-cell RBF
scheme as
D =
_
(d
1
, d
2
, , d
C
) R
C
+
: (
1
,
2
, ,
C
) R
C
+
C

c=1

c
d
c
lim

_
max
0McNT
C

c=1

c
R
(c)
sum
log
2

__
,
(4)
where = P
T
/
2
denotes the SNR.
In this paper, we are interested in characterizing the
above DoF region for the regime with very large num-
ber of users in each cell. Specically, we assume that
the number of users in each cell scales with in the or-
der of
c
, with
c
0, denoted by K
c
= (
c
), i.e.,
K
c
/
c
a as with a being a positive constant.
3. SINGLE-CELL CASE
In this section, we investigate the achievable DoF of RBF
in a simple single-cell case. In this case, the DoF region
in (4) collapses to a line, bounded by 0 and d

, where d

denotes the maximum DoF available for a single cell. For

brevity, we drop the cell index c in this section. Thus,
(2) reduces to
SINR
k,m
=
PT
M
|h
k

m
|
2

2
+
PT
M

M
i=1,i =m
|h
k

i
|
2
. (5)
The CDF of SINR
k,m
, k, m can be expressed as [2]
F
S
(s) = 1
e
s/
(s + 1)
M1
, (6)
where = P
T
/(M
2
) is the SNR per beam. We dene
the achievable DoF for a given and M as
d(, M) = lim

R
sum
log
2

= lim

R
sum
log
2

. (7)
Therefore, for a given , the maximum DoF is obtained
as d

() = max
0MNT
d(, M).
Lemma 3.1. Assume that K = (

), the DoF of
single-cell RBF with M transmit beams is
d(, M) =
_
M
M1
, 0 M 1,
M, > M 1.
(8)
Proof. Let R
k,1
= log
2
(1 + SINR
k,1
). To show (8), it is
sucient to show (9) and (10) as follows:
Pr
_

M 1
log
2
+ log
2
log max
k=1, ,K
R
k,1

M 1
log
2
log
2
log
_

1, if 0 M 1,
(9)
Pr
_
log
2
+ log
2
log + log
2
max
k=1, ,K
R
k,1
log
2
+ log
2
log + log
2

_

1, if > M 1,
(10)
where, is a constant, 0 < < .
Theorem 3.1. For a single-cell RBF system where the
BS has N
T
antennas and the user density coecient is ,
the maximum achievable DoF and corresponding optimal
number of transmit beams are
1
d

() =

+ 1, NT 1, 1 {}( + 2),
(+2)
+1
, NT 1, {}( + 2) < 1,
NT , > NT 1.
(11)
M

() =

, NT 1, 1 {}( + 2),
+ 1, NT 1, {}( + 2) < 1,
NT , > NT 1.
(12)
In Fig. 1, we use Monte-Carlo simulations to conrm
the validity of Lemma 3.1. It is observed that the sum-
rate scaling law for single-cell RBF is quite accurate,
even for small values of the SNR . We notice that the
sum rate for M = 2 is higher than that of M = 4. This
is because with N
T
= 4 and = 1, the optimal number
of beams to achieve d

= 2 is M

= 2, as shown in Fig.
2. In Fig. 2, we demonstrate the dependence of the
optimal single-cell DoF and number of transmit beams
on the user density . We remark that in a cell where
the number of users is small, e.g., < 1, we only need
to transmit one or two beams to maximize the sum-rate,
which is consistent with observations based on numerical
results in existing literature. To obtain the maximum
available DoF d

= 4 with M
c
= N
T
= 4, the user
density coecient needs to be at least 3.
1
The notations and {} denote the integer and fractional
parts of , respectively.
6 7 8 9 10 11 12 13 14 15 16
10
15
20
25
30
log
2

R
B
F

s
u
m

r
a
t
e

[
b
p
s
/
H
z
]


M=2
M=4
numerical R
sum
d(,M) log
2

Fig. 1. Comparison of the numerical sum rate and the

scaling law d(, M) log
2
, with N
T
= 4, = 1, and
K =

.
0 0.5 1 1.5 2 2.5 3 3.5 4
1
1.5
2
2.5
3
3.5
4



d
*
()
M
*
()
Fig. 2. The maximum DoF d

() and optimal number

of beams M

() with N
T
= 4.
4. MULTI-CELL CASE
In this section, we study the DoF region (4) for the gen-
eral multi-cell case. First, we need to derive the CDF
of SINR
(c)
k,m
given in (2). However, extending the deriva-
tions for the SINR CDF from single-cell to multi-cell
possesses some challenges. To the best of our knowl-
edge, no closed-form expression is available, while some
approximated expressions have been obtained [4]. Thus,
we rst show the following lemma on the SINR CDF in
the multi-cell case
Lemma 4.1. The CDF of SINR
(c)
k,m
, k, m, is
F
(c)
S
(s) = 1
e
s/c
(s + 1)
Mc1
C

l=1,l=c
_

l,c

c
s + 1
_
M
l
. (13)
where
c
= P
T
/(M
c

2
) and
l,c
=
l,c
P
T
/(M
l

2
).
Proof. The proof is omitted due to the space limitation.
We then dene the DoF for the sum-rate of the c-th
cell as d
c
(
c
, M) = lim

R
(c)
sum
log
2

, where M=[M
1
,
,M
C
]
T
, and state the next lemma
Lemma 4.2. Assume that K
c
= (
c
), c = 1, , C,
d
c
(
c
, M) is given by
d
c
(
c
, M) =
_
cMc

C
c=1
Mc1
, 0
c


C
c=1
M
c
1,
M
c
,
c
>

C
c=1
M
c
1.
(14)
Proof. The proof uses Lemma 4.1 and similar arguments
in the proof of Lemma 3.1 , and is thus omitted.
Theorem 4.1. Let =[
1
, ,
C
]
T
and D( , M)=
_
d
1
(
1
, M), , d
C
(
C
, M)

T
. The DoF region of a
C-cell RBF system is characterized as
D( ) = conv
_
D( , M), 0 M
c
N
T
, c = 1, , C
_
.
(15)
Proof. The proof is ommited for brevity.
Esentially, we can obtain the DoF region D( ) by
taking a convex hull operation over all points D( , M)
with dierent values of M.
In Fig. 3, we depict the DoF regions of a two-cell RBF
system with dierent user density coecients
1
and
2
.
The vertices of those regions can be obtained with appro-
priate numbers of beams M
1
and M
2
, while time-sharing
between these vertices yields the entire boundary. Only
when
1
7 and
2
7, the maximum DoF pair, shown
as point P
1
, is achievable with (M
1
, M
2
) = (4, 4). When
the numbers of users in two cells are small, e.g.,
1
= 2
and
2
= 2.5, time-sharing between (M
1
, M
2
)=(3,0) and
(0,4), i.e. each BS transmitting alternately, is the opti-
mal strategy. However, with moderate user density coef-
cients, e.g.
1
= 5.7,
2
= 4.7, orthogonal transmission
of the two BSs is suboptimal, as demonstrated by the
dash-line. In general, the DoF trade-o between the two
cells indicates that a cooperation between the two BSs
is needed.
5. CONCLUSION
This paper studies the asymptotic sum-rates with RBF
in a multi-cell system for the regime of high SNR and
large number of users per cell. The optimal DoF is de-
rived for the single cell case in closed-form expressions,
while a characterization for the achievable DoF region
in the general multi-cell case is presented. Insights for
cooperative BS transmission with RBF are drawn.
0 1 2 3 4
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5

1
=2,
2
=2.5

1
=5.7,
2
=4.7
P
1

1
=7,
2
=7
d
1

d
2
Fig. 3. Achievable DoF region of a two-cell RBF system,
with N
T
= 4.
6. REFERENCES
[1] P. Viswanath, D. N. C. Tse, and R. Laroia, Oppor-
tunistic beamforming using dumb antennas, IEEE
Trans. Inf. Theory, vol. 48, No. 6, pp. 1277-1294,
June 2002.
[2] M. Sharif, and B. Hassibi, On the capacity of
MIMO broadcast channel with partial side informa-
tion, IEEE Trans. Inf. Theory, vol. 51, no. 2, pp.
506-522, Feb. 2005.
[3] J. Wu, Y. Hsieh, and P. Ting, Multicell collabora-
tive Random Beamforming with power sharing and
user selection, in Proc. GLOBECOM 2010, pp. 1-
5, 2010.
[4] S.-H. Moon, S.-R. Lee, and I. Lee, Sum-rate ca-
pacity of random beamforming for multi-antenna
broadcast channels with other cell interference,
IEEE Trans. Wireless Commun., Aug. 2011.
[5] V. R. Cadambe, and S. A. Jafar, Interference align-
ment and degrees of freedom of the K-user interfer-
ence channel, IEEE Trans. Inf. Theory, vol. 54, pp.
3425-3441, Aug. 2008.
[6] S.-H. Park, and I. Lee, Degrees of freedom of mul-
tiple broadcast channels in the presence of inter-cell
interference, IEEE Trans. Commun., vol.59, no.5,
pp.1481-1487, May 2011.